📊 FigTabMiner Extraction Results

Caption: Figure 3. SEM Images of the synthesised Iron oxide nanoparticles with reaction times A) 12 hrs, and B) 24hrs at 170°C.

Caption: Figure 3. SEM Images of the synthesised Iron oxide nanoparticles with reaction times A) 12 hrs, and B) 24hrs at 170°C.

Caption: Fig.3).

Caption: Figure (1) Schematics of the ten aromatic molecules used in this study.

Caption: Figure (2) Testing performance for the 10-molecule model. εML is computed by averaging ε predictions from the 10-fold cross-validation. εQC is the quantum mechanically computed excitation energies for the first state.

Caption: Figure (3) Testing performance for the 7-molecule model. εML is computed by averaging ε predictions from the 10-fold cross-validation. εQC is the quantum mechanically computed excitation energies for the first state. Molecules used in model generalization are indicated with blue panels.

Caption: Figure (4) UV-Visible absorption spectra for all 10 aromatic molecules. Thick lines repre- sent the ML spectra predicted using the 10-molecule model and computed with the ensemble method. Thin lines represent the experimental reference.6,72–76 Dashed lines represent the calculated spectra using the multiscale quantum chemical method.

Caption: Figure (5) UV-Visible absorption spectra for all 10 aromatic molecules. Thick lines rep- resent the ML spectra computed with the ensemble method (dashed) and with the third order cumulant scheme (solid).21 Thin lines represent the experimental references.6,72–76 The spectra of the molecules not included in training set are highlighted with blue graph frames and labels.

Caption: Figure (6) Linear decomposition analysis from the 7-molecule model. %group are computed by averaging the excitation energy predictions of 5000 frames.

Caption: Figure (7) Testing performance for the 10-molecule model. εML is computed by averaging ε predictions from the 10-fold cross-validation. εQC is the quantum mechanically computed excitation energies for the second state.

Caption: Figure (8) UV-Visible absorption spectra for all 10 aromatic molecules, including the first two excited states. Thick lines correspond to the ML spectra predicted using the 10-molecule model. Dotted lines represent the calculated spectra using the multiscale quantum chemical method.

Caption: No caption found.

Caption: Figure (2) Testing performance for the 10-molecule model. εML is computed by averaging ε predictions from the 10-fold cross-validation. εQC is the quantum mechanically computed excitation energies for the first state.

Caption: Figure (3) Testing performance for the 7-molecule model. εML is computed by averaging ε predictions from the 10-fold cross-validation. εQC is the quantum mechanically computed excitation energies for the first state. Molecules used in model generalization are indicated with blue panels.

Caption: Figure (4) UV-Visible absorption spectra for all 10 aromatic molecules. Thick lines repre- sent the ML spectra predicted using the 10-molecule model and computed with the ensemble method. Thin lines represent the experimental reference.6,72–76 Dashed lines represent the calculated spectra using the multiscale quantum chemical method.

Caption: Figure (5) UV-Visible absorption spectra for all 10 aromatic molecules. Thick lines rep- resent the ML spectra computed with the ensemble method (dashed) and with the third order cumulant scheme (solid).21 Thin lines represent the experimental references.6,72–76 The spectra of the molecules not included in training set are highlighted with blue graph frames and labels.

Caption: Figure (6) Linear decomposition analysis from the 7-molecule model. %group are computed by averaging the excitation energy predictions of 5000 frames.

Caption: Figure (7) Testing performance for the 10-molecule model. εML is computed by averaging ε predictions from the 10-fold cross-validation. εQC is the quantum mechanically computed excitation energies for the second state.

Caption: Figure (7) Testing performance for the 10-molecule model. εML is computed by averaging ε predictions from the 10-fold cross-validation. εQC is the quantum mechanically computed excitation energies for the second state.

Caption: No caption found.

Caption: FIG. 2. Density plot showing the distance upper limit for GW detectors (dUL GW) (see Eq. 5) for CE (left), ET (middle), and ET+CE (right) on the δt - fth plane.

Caption: FIG. 2. Density plot showing the distance upper limit for GW detectors (dUL GW) (see Eq. 5) for CE (left), ET (middle), and ET+CE (right) on the δt - fth plane.

Caption: FIG. 2. Density plot showing the distance upper limit for GW detectors (dUL GW) (see Eq. 5) for CE (left), ET (middle), and ET+CE (right) on the δt - fth plane.

Caption: FIG. 2. Density plot showing the distance upper limit for GW detectors (dUL GW) (see Eq. 5) for CE (left), ET (middle), and ET+CE (right) on the δt - fth plane.

Caption: FIG. 2. Density plot showing the distance upper limit for GW detectors (dUL GW) (see Eq. 5) for CE (left), ET (middle), and ET+CE (right) on the δt - fth plane.

Caption: FIG. 2. Density plot showing the distance upper limit for GW detectors (dUL GW) (see Eq. 5) for CE (left), ET (middle), and ET+CE (right) on the δt - fth plane.

Caption: Figure 4 Schematics drawings of emitted electrons contributing to different contrasts. (a) Energy distribution of

Caption: No matching caption found.

Caption: No matching caption found.

Caption: No matching caption found.

Caption: Figure 5 Contrast separation by changing the WD. (a) Sketch drawing of the

Caption: Figure 5 Contrast separation by changing the WD. (a) Sketch drawing of the

Caption: Figure 5 Contrast separation by changing the WD. (a) Sketch drawing of the

Caption: Figure 6 Contrast separation by changing the deceleration voltage in the Hitachi

Caption: Figure 6 Contrast separation by changing the deceleration voltage in the Hitachi

Caption: Figure 6 Contrast separation by changing the deceleration voltage in the Hitachi

Caption: Figure 6 Contrast separation by changing the deceleration voltage in the Hitachi

Caption: Figure 6 Contrast separation by changing the deceleration voltage in the Hitachi

Caption: Figure 6 Contrast separation by changing the deceleration voltage in the Hitachi

Caption: No matching caption found.

Caption: No matching caption found.

Caption: Figure 7 Contrast evolution due to the electron-beam-induced deposition in LVSEM images of polymer-sorted

Caption: Figure 8 LVSEM contrast separation of a CVD-grown CNT using a Hitachi Regulus

Caption: Figure 9 LVSEM and cross-section STEM images of a polymer-sorted CNT array. (a) LVSEM image obtained

Caption: Figure 9 LVSEM and cross-section STEM images of a polymer-sorted CNT array. (a) LVSEM image obtained

Caption: Figure 9 LVSEM and cross-section STEM images of a polymer-sorted CNT array. (a) LVSEM image obtained

Caption: Figure S3 The acceptance map of the Inlens detector at a WD of (a) 1 mm, (b) 3 mm, and (c) 5 mm.

Caption: Figure S3 The acceptance map of the Inlens detector at a WD of (a) 1 mm, (b) 3 mm, and (c) 5 mm.

Caption: Figure S4 SEM images of an uncleaned polymer-sorted CNT array. (a) Before and (b) after annealing in Ar gas at 350 ℃ for 1 hour. A WD of 5 mm is used to emphasize the material difference. The yellow arrow indicates the contrast referece, i.e. bright contrast of the bare substrate near the edge of the CNT arrays. White spots circled in red indicate the partial removal of the covered polymers. (c) θ

Caption: Figure S4 SEM images of an uncleaned polymer-sorted CNT array. (a) Before and (b) after annealing in Ar gas at 350 ℃ for 1 hour. A WD of 5 mm is used to emphasize the material difference. The yellow arrow indicates the contrast referece, i.e. bright contrast of the bare substrate near the edge of the CNT arrays. White spots circled in red indicate the partial removal of the covered polymers. (c) θ

Caption: Figure S5 The acceptance map of the Upper and Top detectors at a Vdec of (a) 0, (b) -100 V, and (c) -

Caption: Figure S5 The acceptance map of the Upper and Top detectors at a Vdec of (a) 0, (b) -100 V, and (c) -

Caption: Figure S5 The acceptance map of the Upper and Top detectors at a Vdec of (a) 0, (b) -100 V, and (c) -

Caption: Figure S6 Statistics of the number of CNTs in 1 μm in the typical region. (a) A LVSEM image

Caption: Table 1 Imaging parameters of the two SEM instruments. The beam current is measured by a Faraday

Caption: Figure 2 Schematics illustrating the charge contrast of a single CNT, a low-density CNT array (pitch> 30 nm),

Caption: Figure 2 Schematics illustrating the charge contrast of a single CNT, a low-density CNT array (pitch> 30 nm),

Caption: Figure 2 Schematics illustrating the charge contrast of a single CNT, a low-density CNT array (pitch> 30 nm),

Caption: Figure 4 Schematics drawings of emitted electrons contributing to different contrasts. (a) Energy distribution of

Caption: Figure 7 Contrast evolution due to the electron-beam-induced deposition in LVSEM images of polymer-sorted

Caption: Fig. 1: Overview of the training pipeline for structure elucidation. Characteristic spectral peaks are extracted from raw IR, Raman, UV, NMR, or MS data and used to construct natural language prompts. These are input to a frozen large language model fine-tuned via LoRA. The model is trained to autoregressively generate molecular structures in SMILES format, supervised by the ground-truth sequence.

Caption: Fig. 2: Effect of Raman spectra on structural prediction accuracy. Three representative examples where incorporating Raman spectra corrects wrong predictions made using only IR or UV-Vis inputs. This highlights Raman’s complementary role in resolving molecular substructures sensitive to polarizability.

Caption: Fig. 3: Importance of IR spectra for identifying functional groups. Three representa- tive examples where IR spectra are essential to correctly identify carbonyl groups and distinguish branched chain configurations. Without IR input, predictions based on Raman and UV-Vis remain ambiguous or incorrect.

Caption: Table 6: Data distribution across spectral modalities and splits.